Skip to content

hisilicon-ev200 / goke-gk7205v200: IMX335 boost-1944p latency framing#2091

Merged
widgetii merged 1 commit into
masterfrom
imx335-boost-latency-doc-link
May 13, 2026
Merged

hisilicon-ev200 / goke-gk7205v200: IMX335 boost-1944p latency framing#2091
widgetii merged 1 commit into
masterfrom
imx335-boost-latency-doc-link

Conversation

@widgetii
Copy link
Copy Markdown
Member

Summary

Rewrites the IMX335 boost-1944p INI header comment to describe what the mode is actually for.

The encoder caps wire-rate at ~21 fps for full-5M h.264 4 Mbps regardless of video0.size, so the wire-fps number alone makes this preset look pointless next to stock 5M. The real benefit is glass-to-glass latency:

  • sensor readout per frame: 25 ms (vs 33 ms stock 5M)
  • encoder picks from a 40 fps stream instead of 30 fps (~4 ms fresher on average)
  • combined: ~16 ms shorter glass-to-glass latency at identical wire fps

That 16 ms claim is analytical, not measured. Adds a link from the INI header to the new methodology spec for verifying it (in flight at OpenIPC/openhisilicon#103) so the next person can run actual numbers.

Also expands the header with the three-line enable recipe (sensorConfig + video0.size=2592x1944 + video0.fps=45) so users don't accidentally downscale to 1920×1080 and get a 4:3-cropped image.

Companion to OpenIPC/openhisilicon#103. Independent of it — INI just adds a comment-only link.

No register/INI value changes. Header text only.

Test plan

  • Files install at /etc/sensors/high-fps/imx335_2592x1944_45fps.ini on both ev200 and gk7205v200 firmware images (no install-rule changes; pure file edit)
  • cli/majestic parses the INI unchanged (header is ;-comment-only; no register/section edits)
  • Boost preset still delivers identical sensor + encoder behavior as PR hisilicon-ev200 / goke-gk7205v200: IMX335 companion to openhisilicon #99 #2090

🤖 Generated with Claude Code

Rewrites the boost preset's INI header to describe what the mode is
actually for: glass-to-glass latency, not wire fps. The encoder
caps at ~21 fps for full-5M h.264 4 Mbps regardless of video0.size,
so the wire-fps number alone makes this preset look pointless next
to stock 5M. The real win is ~16 ms shorter glass-to-glass latency
(sensor readout 25 ms vs 33 ms, plus encoder picks from a 40 fps
stream instead of 30 fps).

That 16 ms claim is analytical, not measured. Adds a link to the
methodology spec for verifying it:
openhisilicon/docs/sensor-pipeline-latency-metering.md
(in flight at OpenIPC/openhisilicon#103).

Also expands the header with the three-line enable recipe
(sensorConfig + video0.size + video0.fps) so users don't try to
downscale to 1920x1080 and get a 4:3-cropped image.

No INI register changes, no functional change.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@widgetii widgetii merged commit 4e0e173 into master May 13, 2026
91 checks passed
@widgetii widgetii deleted the imx335-boost-latency-doc-link branch May 13, 2026 09:06
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant